nfw2

The way it understands it is the same way it understands anything else, by reasoning about it in legible human terms. It's not that surprising that these LLMs, trained to be effective in technical fields, are good at the technical field of understanding the implications of model weights.

It's not like these models understand weights be some special linear algebra communion with its own weights.

One could say they might be "superhuman" here in the same way that LLMs might be said to be superhuman at math. Perhaps they do this work faster and more effectively than any person could manually. But they are certainly still doing it the "human" way ie reasoning with legible and coherent thoughts, rather communing through some Matrix-like symbology which is what people who skimmed the article seem to be assuming.

Growth is first derivative. Deceleration is second derivative. Focusing on second not first I think was the point, not implying there is some nth derivative under discussion.

The mistake is not understanding that when code becomes a commodity, it's time to move up to the business chain and find problems to own.

All the incumbents should be afraid of pissed off engineers trying to get their business or building custom stacks inside their biggest clients. Screaming into the void by releasing more code isn't going to move the needle.

I don't think pets.com, an online retailer, is an apt comparison to Anthropic, an bleeding edge research lab doing things that most people would consider science fiction a few years ago.

This is what the parent post said:

ship fast go-to market features in the blink of an eye when it becomes necessary

Necessary go-to market features are useful

How is shipping features fast not productivity?

That's still increased productivity even if the consumer benefits instead of the business owner.

If you're getting more done but unsure of the resulting value, that is indicative of your role not the tools you're using.

I am telling what's on Anthropic's official announcement. It's not a subjective assessment.

"Inching closer", yes. "Soon" seems presumptive. FSD everywhere seemed "soon" 20 years ago.

explains how fringe details don't make sense, like no one knowing what oracle does exactly

One can't be both earnest and cynical.

1. In Anthropic's benchmarks, Opus's advantage over Sonnet diminishes at high reasoning levels. Sonnet max actually beats Opus max.

2. There are rote or simple tasks where higher reasoning isn't needed and one model vs the other wouldn't even be distinguishable.

3. How many tokens are used during reasoning depends as much on the task as the reasoning level.

4. Sonnet is faster and half the price per token, so Opus will only be cheaper if it can spend fewer tokens. This won't be the case for all tasks. Opus and Sonnet will also consume the same number of tokens to load context, so the relative price comparison depends a lot on where the task is context-heavy or reasoning-heavy.

There is a chasm of performance difference between haiku and sonnet ime. Sonnet is cheaper and faster than Opus, so if you're providing a service via api where cost and speed is important, it's a big difference.

The claim that is being made in the article is that only cultural differences are relevant. Saying the relative effects of cultural and biological differences are hard to know would not be a very controversial statement.

This is the paragraph. "As an example" here implies that they are talking about sports in general, not just endurance events.

Inequity between male and female athletes is a result not of inherent biological differences between the sexes but of biases in how they are treated in sports. As an example, some endurance-running events allow the use of professional runners called pacesetters to help competitors perform their best...

The worst thing about this article is it seems to fail to understand the basic of hypothesis-testing, the bedrock of science.

Title: The Theory That Men Evolved to Hunt and Women Evolved to Gather Is Wrong

Subtitle: The influential idea that in the past men were hunters and women were not isn’t supported by the available evidence

Saying there isn't enough evidence to prove a hypothesis is not the same as proving its negation. Failing to reject the null hypothesis is not the same as proving it.

It seems to me that hype cycle of jev made it harder to understand rather than easier.

Universal classifier just seems nonsensical. It seems clear that it can't have accurate probability distributions on any possible question you might ask it. Moreover, giving only "maybe" answers to everything adds a layer of equivocation that makes it harder to catch it in a clear mistake.

That said, I've encountered plenty of fuzzy judgment questions now (ie is this case more like example X or example Y?) that reasoning models struggle with. I could see how, in lieu of a dedicated classifier, a model like jev, ostensibly trained to guesstimate the relative likelihood of natural language hypotheses, may do well.

But it still feels wrong to call it a classifier in the traditional sense, let alone a universal classifier.

It isn't a cynical scam to make money. They are genuinely stupid.

First I agree RFK jr is the worst possible HHS secretary imaginable. There has been tons of funding pulled for mrna vaccine research for respiratory viruses which is terrible.

That said:

1. Funding hasn't been pulled for mrna vaccines for cancer. rfk has been equivocal about the topic. NCI is stilling rewarding grants for mrna-vaccine cancer trials.

2. Pulling funding is different from blocking vaccines from reaching the market. The new season's covid vaccine is presumably an mrna vaccine. For cancer vaccines, it would be very strange to fund a vaccine just to block it from the market.

Should we hold these people accountable?

What do you mean? The court of public opinion seems to hold them accountable. Are you proposing legal punishments for Marc Andreessen? For what specifically?

I think the problem is moreso that making useful things for enterprise with AI right now is actually pretty hard because context and attention management is hard, the token cost for complex tasks scales fast, and model's lack of intuition for fuzzy decision-making.

There are hordes of grifters promising things they can't deliver and reaping in millions in capital from investors who are equally confused about the challenges of designing useful AI.

That said, AI can do useful things when properly designed. The market finding the people who can make this happen won't happen as fast as people expect.

> Adopting AI tools that somebody else created, and are equally available to your competitors, won't increase your revenue or margins

The only way more efficient operations doesn't make margins go up is if prices go down. Is that what we're seeing?

having an index of the code that the agent can use to find what its looking for instead of grep seems to work wonders

maybe they are doing the fast food pay it forward trick from the tim robinson show. someone else adds more, then they take it

https://www.youtube.com/watch?v=52LJaWDdG9c

> You have better than zero in a trillion odds against Lebron James. What if his phone rings?

You can easily devise in this thought experiment rules that prevent loopholes, like saying no phones allowed and medical emergencies do not result in a win for the opponent.

The award was for the illustration, it wasn't incidental.

> need a way to filter what's actually worth investing time and attention to

nothing on social media is worth investing time and attention to. including hacker news, yet here i am so i guess my experience is not that different from the average experience, it's just not because of ai

> social media of choice

tiktok, x, linkedin, etc. have been garbage as long as they have existed. if ai slop made tiktok worse i would welcome that as a good thing for society.

> search for something online

just search on claude. the water is warm here on the dark side

> low effort drama in streaming sites

these are front and center because people like them. they have filters for acclaimed movies that aren't front and center because people don't want to watch There Will Be Blood on a random Tuesday after work. this seems like a reach

In your understudy metaphor, you seem to be suggesting that flaws with AI art are fundamentally due to deficient craft.

In that case, if you saw or read something that you considered high quality, and later found out it was AI, would that affect your opinion of it?

A contest of skill is not necessarily like spinning a roulette.

If I were to play Lebron James in basketball 1v1, it would not matter how many times we ran the experiment. We could play a trillion times, and I would win 0 of them.

If you sampled 10 Hollywood directors and asked whether The Godfather or The Room is a better movie, the Room would win the poll 0 out of a trillion times.

The reasonable naive expectation for contests like these is that, below a certain bar of quality, the chance of win is 0. So if you see wins go up as volume of submissions goes up, it shows that either ai writing is above that bar, or that faulty judging is a widespread problem beyond a few anomalous organizations.